Unifying User and Message Clustering Information for Retweeting Behavior Prediction

Abstract

Online social networks have been recently increasingly become the dominant platform of information diffusion by user’s retweeting behavior. Thus, understanding and predicting who will be retweeted in a given network is a challenging but important task. Existing studies only investigate individual user and message for retweeting prediction. However, social influence and selection lead to formation of groups. The intrinsic and important factor has been neglected for this problem. In the paper, we propose a unified user and message clustering based approach for retweeting behavior prediction. We first cluster users and messages into different groups based on explicit and implicit factors together. Then we model social clustering information as regularization terms to introduce the retweeting prediction framework in order to reduce sparsity of data and improve accuracy of prediction. Finally, we employ matrix factorization method to predict user’s retweeting behavior. The experimental results on a real-world dataset demonstrate that our proposed method effectively increases accuracy of retweeting behavior prediction compared to state-of-the-art methods.